AI-Based Smart Energy Management
Detecting idle power waste
A machine learning project focused on detecting appliance states and predicting energy consumption patterns to reduce idle power waste.
Project stack
Visual proof
AI-Based Smart Energy Management
This project is presented through its visual identity, stack, highlights and case-study notes, with the strongest screen-level walkthroughs reserved for the featured mobile apps.
Case study
problem
Idle and standby appliances can waste power quietly across homes, offices and institutions.
built
I experimented with appliance-state classification and energy prediction using Python, pandas, scikit-learn, engineered features, baselines and time-series validation.
challenge
Energy datasets can produce misleadingly high scores if splits are too easy, so baseline comparison and validation awareness were important.
learned
I learned how electrical engineering context can guide AI feature engineering, model evaluation and practical energy optimization thinking.
impact
The project shows research-oriented AI/ML ability connected to a real engineering problem instead of a generic model demo.
Overview
The classification work used engineered power, lag, rolling-window and grouped sequence features to detect OFF, IDLE and ACTIVE appliance states.
The regression side used chronological splitting, time features, lags, rolling averages and a HistGradientBoostingRegressor to predict appliance energy consumption.
Highlights
Explored FIRED, LIT, BLOND, UK-DALE, REDD, Pecan Street and ECO dataset directions.
Reported balanced accuracy around 0.9983 and macro-F1 around 0.9820 in one classification run, compared against a 0.3333 baseline.
Reported test MAE around 28.48 against a baseline MAE around 52.53, with R2 around 0.5554 for regression.
Handled results carefully with baseline and validation awareness to reduce leakage risk.